Ensemble Machine Learning Approach Improves Predicted Spatial Variation of Surface Soil Organic Carbon Stocks in Data-Limited Northern Circumpolar Region

Abstract

Various approaches of differing mathematical complexities are being applied for spatial prediction of soil properties. Regression kriging is a widely used hybrid approach of spatial variation that combines correlation between soil properties and environmental factors with spatial autocorrelation between soil observations. In this study, we compared four machine learning approaches (gradient boosting machine, multinarrative adaptive regression spline, random forest, and support vector machine) with regression kriging to predict the spatial variation of surface (0-30 cm) soil organic carbon (SOC) stocks at 250-m spatial resolution across the northern circumpolar permafrost region. We combined 2,374 soil profile observations (calibration datasets) with georeferenced datasets of environmental factors (climate, topography, land cover, bedrock geology, and soil types) to predict the spatial variation of surface SOC stocks. We evaluated the prediction accuracy at randomly selected sites (validation datasets) across the study area. We found that different techniques inferred different numbers of environmental factors and their relative importance for prediction of SOC stocks. Regression kriging produced lower prediction errors in comparison to multinarrative adaptive regression spline and support vector machine, and comparable prediction accuracy to gradient boosting machine and random forest. However, the ensemble median prediction of SOC stocks obtained from all four machine learning techniques showed highest prediction accuracy. Although the use of different approaches in spatial prediction of soil properties will depend on the availability of soil and environmental datasets and computational resources, we conclude that the ensemble median prediction obtained from multiple machine learning approaches provides greater spatial details and produces the highest prediction accuracy. Thus an ensemble prediction approach can be a better choice than any single prediction technique for predicting the spatial variation of SOC stocks.

Large Soil Carbon Storage in Terrestrial Ecosystems of Canada

“…This is consistent with Mishra et al. (2020, 2021) who reported surface temperature and elevation as the main predictors of SOC stocks in permafrost regions, and Guevara et al. (2020), who reported temperature and topographic variables among the best predictors of SOC stock in Latin America.…”

AI-Constrained Bottom-Up Ecohydrology and Improved Prediction of Seasonal, Interannual, and Decadal Flood and Drought Risks

“…Since soil bulk properties, and soil organic carbon in particular, control soil moisture and evapotranspiration, water cycle predictions are strongly dependent on correctly characterizing those properties in models. Recent studies show that AI/ML approaches show promise in improving spatial prediction of soil and environmental properties, including soil organic carbon stocks and land surface fluxes, with field data (Mishra et al, 2020). ML approaches have also been used to derive scaling functions of soil and ecosystem properties (Adhikari et al, 2020).…”

Continental United States may lose 1.8 petagrams of soil organic carbon under climate change by 2100

“…In contrast to previous studies, our approach uses (a) the most recently available SOC stock field observation data from the United States, (b) a large set of environmental controller data, and (c) an ensemble ML approach that has been documented to produce more accurate results (Mishra et al, 2020). ML methods provide an opportunity to improve our understanding of SOC stocks by using high resolution data and to better understand and address the limitations of existing models that rely on a substantial number of input parameters with user-defined values.…”

“…Earlier studies estimated continental US SOC stocks based on mean SOC stock from the State Soil Geographic (STATSGO) database (Guo et al., 2006) and a geographically weighted regression of observations from the Rapid Carbon Assessment database (Gonçalves et al., 2021; West et al., 2013). In contrast to previous studies, our approach uses (a) the most recently available SOC stock field observation data from the United States, (b) a large set of environmental controller data, and (c) an ensemble ML approach that has been documented to produce more accurate results (Mishra et al., 2020). ML methods provide an opportunity to improve our understanding of SOC stocks by using high resolution data and to better understand and address the limitations of existing models that rely on a substantial number of input parameters with user‐defined values.…”

“…Field observations of SOC and data on environmental controllers are valuable for quantifying SOC change and validating model predictions. ML approaches can use these field observations as training data to develop both linear and nonlinear relationships among a large number of environmental factors and SOC observations for better predictions of SOC stocks and thus can help to benchmark environmental control representation in ESMs (Mishra et al., 2020). Earlier studies estimated continental US SOC stocks based on mean SOC stock from the State Soil Geographic (STATSGO) database (Guo et al., 2006) and a geographically weighted regression of observations from the Rapid Carbon Assessment database (Gonçalves et al., 2021; West et al., 2013).…”